Multi-algorithm coordinated public security risk prediction method based on space-time and video perception data

Abstract

This invention relates to the field of public safety risk management technology, and in particular to a multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data. The method includes: obtaining historical risk warning event data from a database, preprocessing it to obtain a historical warning event dataset; extracting spatiotemporal features of warning events from the historical warning event dataset; using the spatiotemporal features of warning events to predict the probability of a risk event occurring in a target area at a future time; and guiding the real-time acquisition of video data in the target area and the prediction of public safety risks based on the probability and historical warning features.

Description

Technical Field

[0001] This invention relates to the field of public safety risk management technology, and in particular to a multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data. Background Technology

[0002] Public safety risk prediction is a complex systems engineering project. Its core is to use multiple algorithms to intelligently integrate and analyze data in order to identify safety risks in advance.

[0003] In reality, video image data is typically collected through public security cameras. By analyzing these images, dangerous behaviors are identified and warnings are issued. However, areas without camera coverage cannot be used for risk prediction based on video data, becoming blind spots for safety risk warnings.

[0004] Furthermore, indiscriminate video surveillance across different security monitoring scenarios results in massive amounts of video stream data, putting pressure on data storage and processing. To improve the ability to identify and predict risks in specific scenarios, targeted data acquisition and processing are necessary, followed by accurate analysis using appropriate algorithm models.

[0005] To address this, we propose a multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data. Summary of the Invention

[0006] This invention obtains the spatiotemporal characteristics of early warning events in a target monitoring area by using historical early warning event data, and uses Hawkes processes to predict the probability of occurrence of corresponding types of early warning events in the future, thereby enabling the prediction of security risks in the corresponding area without video data.

[0007] The technical solution proposed in this invention is: a multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data, the method comprising:

[0008] Historical risk warning event data is obtained from the database and preprocessed to obtain a historical warning event dataset.

[0009] Extracting the spatiotemporal features of early warning events from historical early warning event datasets;

[0010] By utilizing the spatiotemporal characteristics of early warning events, the probability of a risk event occurring in a target area at a future time can be predicted.

[0011] Based on probability and historical early warning characteristics, it guides the real-time acquisition of video data in target areas and the prediction of public safety risks.

[0012] Preferably, the step of extracting the spatiotemporal features of early warning events from the historical early warning event dataset includes:

[0013] Extract the occurrence time, three-dimensional spatial coordinates, type, severity score, and duration of historical early warning events from the historical early warning event dataset to construct the spatiotemporal feature vector of the early warning events. ;

[0014] in, Indicates the first One warning event; , , , , Indicates the first The time, location, type, severity score, and duration of each warning event;

[0015] , ,in, Indicates the first The three-dimensional coordinates of the location where each warning event occurred.

[0016] Preferably, the step of using the spatiotemporal characteristics of the early warning event to predict the probability of a risk event occurring in the target area at a future time includes:

[0017] Obtain the spatiotemporal feature vector of the early warning event Constructing a spatiotemporal sequence of historical early warning events , Indicates the number of warning events;

[0018] Warning events for a future time. The conditional intensity function of the warning event is constructed using a Hawkes process. ;in, Indicates at time ,Location Location, type The probability of a warning event occurring;

[0019] base strength ,in, Indicates the basic strength weight;

[0020] Spatiotemporal basic features ;in, Indicates position Two-dimensional coordinates;

[0021] Trigger function ;

[0022] Among them, time-triggered core , Indicates the time decay coefficient;

[0023] Space Trigger Core covariance matrix It is used to control the spatial influence range and directionality; Indicates a warning event Within the affected area The variance of coordinates Indicates a warning event Within the affected area The variance of coordinates Represents the covariance coefficient;

[0024] in, Indicates the type interaction coefficient;

[0025] Severity adjustment items ,in, Indicates the severity amplification factor. Indicates an event The severity.

[0026] Construct the likelihood function and solve for the parameters within the conditional strength function.

[0027] Preferably, the construction of the likelihood function and the solution of the parameters within the conditional strength function include:

[0028] Construct the likelihood function: ;in, Indicates the scope of the target area;

[0029] The likelihood function is solved using the gradient descent algorithm to obtain the likelihood parameters. .

[0030] Preferred options also include:

[0031] For each historical warning event, Hawkes features are extracted, including:

[0032] Background contribution ;

[0033] Self-triggered contribution

[0034] Interactive trigger contribution ;

[0035] Target Area Early Warning Event Predicted trigger number ;

[0036] Characteristics of the scope of impact of early warning events ;

[0037] The above features are combined to obtain the Hawkes feature vector. ;

[0038] For each historical early warning event, event-level features are extracted using a recurrent neural network (RNN), including:

[0039] Encode each historical early warning event to obtain an event encoding vector. Among them, time coding ; Indicates the first The frequency of historical occurrences; Indicates type encoding;

[0040] Periodic coding ;

[0041] Spatial coding ;in, This represents a multilayer perceptron;

[0042] Each historical early warning event is sorted by time and divided into time windows to obtain the spatiotemporal sequence of historical early warning events. ; Indicates the length of the time window; Indicates the number of time windows;

[0043] The bidirectional long short-term memory (LSTM) algorithm is used to extract RNN feature vectors from the spatiotemporal sequence of historical early warning events. Among them, the bidirectional hidden state vector Hidden state one Hidden state two ;

[0044] After aligning the Hawkes feature vector and the RNN feature vector on the time axis, they are then projected to a unified dimension to obtain the Hawkes feature vector with a unified spatiotemporal dimension. and RNN feature vectors ;

[0045] in, Represents the Hawkes eigenmap matrix. Represents the feature mapping matrix of an RNN. and Represents the Hawkes feature correction vector and the RNN feature correction vector;

[0046] use and Obtain a risk score for the target area at future moments.

[0047] Preferably, the utilization and Obtain a risk score for the target area at future times, including:

[0048] Through multiple algorithms and Integration, to obtain the corresponding integration features, including:

[0049] By splicing, splicing integration features are obtained. ;

[0050] By using gating fusion, we can obtain gating integration features. ;in, Indicates trainable gating parameters;

[0051] Using integrated features to predict the risk of a target region includes:

[0052] Divide the target area into multiple grids. , ,in, Indicates the number of grid cells;

[0053] Predicting future time-point grid risk scores using ensemble vectors or , Represents the weight vector. Indicates the correction amount.

[0054] Preferred options also include:

[0055] The method of guiding real-time acquisition of video data and prediction of public safety risks in target areas based on probability and historical early warning characteristics includes:

[0056] Guided real-time acquisition of video data for the target area, including:

[0057] Based on the conditional intensity function, a spatial risk heatmap of the target area is defined: Weight value ;

[0058] Based on RNN feature vectors, the risk time of the target region is defined. , Represents the Fast Fourier Transform. Indicates attention weights. Represents the Dirac function;

[0059] Calculate the spatial coverage priority of the camera, including:

[0060] For cameras Let its coverage area be and ,in, Indicates the target area;

[0061] Then the camera Spatial coverage priority ;in, For camera In position Observation quality weights;

[0062] Calculate camera time scheduling priority , This indicates the calculation of the correlation coefficient. Indicates camera Historically, periods when high-risk events were captured on film;

[0063] Set and adjust priorities for different event types. ;

[0064] Then, the overall priority function ;in, , , Indicates priority weight. Indicates the event type weight;

[0065] Based on the characteristics of the impact range of the early warning event, obtain the camera... optimal perspective ; Indicates a Gaussian distribution. , These represent the mean and variance of the adjustment of the camera's field of view relative to its original position, respectively.

[0066] Based on the overall priority of the cameras, the corresponding cameras are scheduled first and their vision is adjusted to the optimal angle to collect video data of their coverage area in real time; the video data collected by multiple cameras constitutes the video stream data of the target area.

[0067] Based on the acquired video stream data of the target area, public safety risk prediction is performed, including:

[0068] Construct type interaction matrix Feature extractors are selected based on type interaction matrices, including:

[0069] ;in, Indicates the threshold of clustering features. Individual interaction feature threshold, Crowd density feature extractor Represents a motion feature extractor. Indicates the target feature extractor. This represents a behavioral feature extractor;

[0070] By combining appropriate feature extractors, relevant features are extracted from video stream data, input into a pre-trained early warning model, and output the probability of security risks.

[0071] Preferably, the step of extracting corresponding features from video stream data through a combination of corresponding feature extractors, inputting them into a pre-trained early warning model, and outputting a security risk probability includes:

[0072] Constructed using 3D convolutional neural networks (I3D) or optical flow networks Extracting motion features of targets from video stream data ;

[0073] The target recognition algorithm YOLO is used to construct Extracting the appearance features of targets from video stream data ;

[0074] The CSRNet network for recognizing crowded scenes is constructed. Extracting crowd density features from video stream data ;

[0075] The behavior of a target in video stream data is encoded using an autoencoder and a memory network, and the behavior anomaly score is output through the memory network. ;

[0076] Will and The input is fed into a pre-trained group risk prediction model, which outputs the group risk probability. ;

[0077] Will and The input is fed into a pre-trained individual risk prediction model, which outputs the probability of individual behavioral risk. ;

[0078] if If so, it is determined that there is a risk to public safety.

[0079] if If so, it is determined that there is an individual safety risk.

[0080] An electronic device for executing the multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data.

[0081] A computer-readable storage medium storing a computer program that is executed by a processor to implement the multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data.

[0082] The beneficial effects of this invention are:

[0083] 1. This invention combines multiple algorithms to predict security risks in areas without video surveillance coverage using historical early warning data through a Hawkes process. Furthermore, it extracts Hawkes features and RNN features from historical early warning events to generate a risk score for the target area at future times, facilitating the implementation of appropriate monitoring measures based on the risk score.

[0084] 2. This invention utilizes historical early warning data to guide video data acquisition and security risk prediction. Specifically, it defines a spatial risk heatmap of the target area based on a conditional intensity function and defines the risk time of the target area based on RNN feature vectors. Based on the risk heatmap, it determines the spatial coverage priority of cameras; based on the risk time, it determines the time-based camera access priority; and it determines adjustment priorities for different event types. Based on the spatial coverage priority, spatial coverage priority, and adjustment priority, it obtains a comprehensive camera priority, thereby enabling the function of calling the corresponding camera to collect video data in the risk area at the risk time for different scenarios.

[0085] 3. This invention utilizes the spatiotemporal characteristics of early warning events to obtain the optimal viewing angle of the camera, thereby ensuring the acquisition of complete video data. After obtaining the corresponding video data, it selects the appropriate feature extractor combination to achieve targeted feature extraction, in order to meet the needs of individual risk prediction and group risk prediction. Attached Figure Description

[0086] Figure 1 This is a flowchart of the multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to the present invention. Detailed Implementation

[0087] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious modifications will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0088] It is understood that the term "a" should be understood as "at least one" or "one or more", that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0089] refer to Figure 1 The technical solution provided by this invention is: a multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data, the method comprising:

[0090] Step 1: Obtain historical risk warning event data from the database, preprocess it to obtain a historical warning event dataset;

[0091] Step 2: Extract the spatiotemporal features of warning events from the historical warning event dataset. This includes the following steps: extracting the occurrence time, three-dimensional spatial coordinates, type, severity score, and duration of historical warning events from the historical warning event dataset to construct a spatiotemporal feature vector for the warning events. ;

[0092] in, Indicates the first One warning event; , , , , Indicates the first The time, location, type (e.g., crowd gathering, chasing, etc.), severity score, and duration of each warning event;

[0093] , ,in, Indicates the first The three-dimensional coordinates of the location where each warning event occurred.

[0094] Step 3: Utilize the spatiotemporal characteristics of the early warning event to predict the probability of a risk event occurring in the target area at a future time. This includes the following steps:

[0095] Step 3.1: Obtain the spatiotemporal feature vector of the early warning event. Constructing a spatiotemporal sequence of historical early warning events , Indicates the number of warning events;

[0096] Step 3.2: Early warning events at a future time. The conditional intensity function of the warning event is constructed using a Hawkes process. ;in, Indicates at time ,Location Location, type The probability of a warning event occurring;

[0097] Among them, basic strength , Indicates the basic strength weight;

[0098] Spatiotemporal basic features ;in, Indicates position Two-dimensional coordinates;

[0099] Trigger function ;

[0100] Time-triggered core , This represents the time decay coefficient, which is greater than 0 and is used to control type [missing information]. The rate at which events decay over time.

[0101] Space Trigger Core In this embodiment, the space triggering core adopts an anisotropic Gaussian core.

[0102] covariance matrix It is used to control the spatial influence range and directionality; Indicates a warning event Within the affected area The variance of coordinates Indicates a warning event Within the affected area The variance of coordinates This represents the covariance coefficient.

[0103] in, Indicates the type interaction coefficient. , used to represent type Warning event triggering type The intensity of the warning event.

[0104] Severity adjustment items ,in, Indicates the severity amplification factor. Indicates an event The severity.

[0105] Step 3.3: Construct the likelihood function and solve for the parameters within the conditional strength function, specifically:

[0106] Construct the likelihood function: ;in, Indicates the target area range.

[0107] The likelihood function is solved using the gradient descent algorithm to obtain the likelihood parameters. .

[0108] The predictions made by the Hawkes process in the above steps are based solely on historical warning event data, outputting the probability of the warning event occurring. It does not rely on video data, and its coverage is not limited by camera deployment, allowing it to cover areas without cameras.

[0109] Step 3.4: Extract Hawkes features for each historical early warning event, including:

[0110] Background contribution ;

[0111] Self-triggered contribution

[0112] Interactive trigger contribution ;

[0113] Target Area Early Warning Event Predicted trigger number ;

[0114] Characteristics of the scope of impact of early warning events ;

[0115] The above features are combined to obtain the Hawkes feature vector. ;

[0116] For each historical early warning event, event-level features are extracted using a recurrent neural network (RNN), including:

[0117] Encode each historical early warning event to obtain an event encoding vector. Among them, time coding ; Indicates the first The frequency of historical occurrences; Indicates type encoding;

[0118] Periodic coding ;

[0119] Spatial coding ;in, This represents a multilayer perceptron;

[0120] Each historical early warning event is sorted by time and divided into time windows to obtain the spatiotemporal sequence of historical early warning events. ; Indicates the length of the time window; Indicates the number of time windows;

[0121] The bidirectional long short-term memory (LSTM) algorithm is used to extract RNN feature vectors from the spatiotemporal sequence of historical early warning events. Among them, the bidirectional hidden state vector Hidden state one Hidden state two ;

[0122] After aligning the Hawkes feature vector and the RNN feature vector on the time axis, they are then projected to a unified dimension to obtain the Hawkes feature vector with a unified spatiotemporal dimension. and RNN feature vectors ;

[0123] in, Represents the Hawkes eigenmap matrix. Represents the feature mapping matrix of an RNN. and This represents the Hawkes feature correction vector and the RNN feature correction vector.

[0124] use and Obtain the risk score for the target area at future moments, specifically:

[0125] Through multiple algorithms and Integration, to obtain the corresponding integration features, including:

[0126] By splicing, splicing integration features are obtained. ;

[0127] By using gating fusion, we can obtain gating integration features. ;in, Indicates trainable gating parameters;

[0128] Using integrated features to predict the risk of a target region includes:

[0129] Divide the target area into multiple grids. , ,in, Indicates the number of grid cells;

[0130] Predicting future time-point grid risk scores using ensemble vectors or , Represents the weight vector. This indicates the correction amount. Different measures are taken based on the risk score, such as increasing manual patrols of the grid or using corresponding cameras to collect video data of the grid area.

[0131] The above steps, through the Hawkes process, provide an interpretable triggering mechanism, facilitating the identification of the causes of warning events. Simultaneously, the use of RNNs helps capture complex sequence patterns among warning events, improving the prediction's fault tolerance. In practical applications, the weight vector can be adjusted to adapt to different scenarios.

[0132] Step 4: Based on probability and historical early warning characteristics, guide the real-time acquisition of video data in the target area and the prediction of public safety risks, specifically including the following steps:

[0133] Step 4.1: Guide the real-time acquisition of video data in the target area, including:

[0134] Step 4.11: Based on the conditional intensity function, define a spatial risk heatmap of the target area to map historical early warning event triggering patterns to video areas of interest, in order to identify high-risk areas. Weight value ;

[0135] Based on RNN feature vectors, a risk time for the target region is defined to map RNN features to the time of interest, thereby identifying high-risk periods.

[0136] Target area risk time , Represents the Fast Fourier Transform. Indicates attention weights. Represents the Dirac function;

[0137] Step 4.12: Calculate the camera spatial coverage priority, including:

[0138] For cameras Let its coverage area be and ,in, Indicates the target area;

[0139] Then the camera Spatial coverage priority ;in, For camera In position Observation quality weights;

[0140] Step 4.13: Calculate camera time scheduling priority , This indicates the calculation of the correlation coefficient. Indicates camera Historically, periods when high-risk events were captured on film;

[0141] Set and adjust priorities for different event types. ;

[0142] Then, the overall priority function ;in, , , The priority weights can be learned from historical data. Indicates the event type weight;

[0143] Step 4.14: Based on the characteristics of the impact range of the early warning event, obtain the camera data. optimal perspective ; Indicates a Gaussian distribution. , These represent the mean and variance of the adjustment of the camera's field of view relative to its original position, respectively.

[0144] Step 4.15: Based on the comprehensive priority of the cameras, prioritize the scheduling of the corresponding cameras and adjust their vision to the optimal angle to collect video data of their coverage area in real time; the video data collected by multiple cameras constitute the video stream data of the target area.

[0145] Step 4.16: Based on the acquired video stream data of the target area, perform public safety risk prediction, including:

[0146] Construct type interaction matrix Feature extractors are selected based on type interaction matrices, including:

[0147] ;in, Indicates the threshold of clustering features. Individual interaction feature threshold, Crowd density feature extractor Represents a motion feature extractor. Indicates the target feature extractor. This refers to a behavioral feature extractor.

[0148] By combining appropriate feature extractors, relevant features are extracted from the video stream data, input into a pre-trained early warning model, and the output is the probability of security risk, specifically:

[0149] Constructed using 3D convolutional neural networks (I3D) or optical flow networks Extracting motion features of targets from video stream data ;

[0150] The target recognition algorithm YOLO is used to construct Extracting the appearance features of targets from video stream data ;

[0151] The CSRNet network for recognizing crowded scenes is constructed. Extracting crowd density features from video stream data ;

[0152] The behavior of a target in video stream data is encoded using an autoencoder and a memory network, and the behavior anomaly score is output through the memory network. ;

[0153] Will and The input is fed into a pre-trained group risk prediction model, which outputs the group risk probability. In this embodiment, the group risk prediction model is built based on a convolutional neural network.

[0154] Will and The input is fed into a pre-trained individual risk prediction model, which outputs the probability of individual behavioral risk. In this embodiment, the individual risk prediction model is constructed based on a recurrent convolutional neural network.

[0155] if If so, it is determined that there is a risk to public safety, such as gatherings or stampedes.

[0156] if If so, it is determined that there is an individual safety risk, such as chasing or fighting.

[0157] This step provides clear guidance for video acquisition based on historical early warning event characteristics, calling on corresponding cameras to collect video data of key grids during key time periods, thereby avoiding blind monitoring and optimizing resource allocation. Simultaneously, by selectively extracting features from the video stream data, and choosing the appropriate feature extractor based on the early warning needs, the efficiency and specificity of prediction are improved; full feature extraction is avoided, reducing the consumption of computational resources.

[0158] The present invention also provides an electronic device for executing the aforementioned multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data.

[0159] The present invention also provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement the multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data.

[0160] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. Computer-readable storage media can be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0161] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0162] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the principles described, the implementation of the present invention may have any changes or modifications.

Claims

1. A multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data, characterized in that, The method includes: Historical risk warning event data is obtained from the database and preprocessed to obtain a historical warning event dataset. Extracting the spatiotemporal features of early warning events from historical early warning event datasets; By utilizing the spatiotemporal characteristics of early warning events, the probability of a risk event occurring in a target area at a future time is predicted; in areas without video surveillance coverage, a Hawkes process is used to predict security risks using historical early warning data; and Hawkes features and RNN features are extracted from historical early warning events to generate a risk score for the target area at a future time. Based on probability and historical early warning characteristics, the system guides real-time acquisition of video data in target areas and prediction of public safety risks. A spatial risk heatmap of the target area is defined based on a conditional intensity function, and the risk time of the target area is defined based on RNN feature vectors. The spatial coverage priority of cameras is determined based on the risk heatmap, the temporal call priority of cameras is determined based on the risk time, and adjustment priorities are determined for different event types. A comprehensive priority for cameras is obtained based on spatial coverage priority, temporal call priority, and adjustment priority. Based on the comprehensive priority of cameras, corresponding cameras are prioritized and their views are adjusted to the optimal perspective to collect video data of their coverage areas in real time. Video data collected by multiple cameras constitutes the video stream data of the target area. Public safety risk prediction is performed based on the acquired video stream data of the target area.

2. The multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to claim 1, characterized in that, The extraction of spatiotemporal features of early warning events from historical early warning event datasets includes: Extract the occurrence time, three-dimensional spatial coordinates, type, severity score, and duration of historical early warning events from the historical early warning event dataset to construct the spatiotemporal feature vector of the early warning events. ; in, Indicates the first One warning event; , , , , Indicates the first The time, location, type, severity score, and duration of each warning event; , ,in, Indicates the first The three-dimensional coordinates of the location where each warning event occurred.

3. The multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to claim 2, characterized in that, The method of predicting the probability of a risk event occurring in a target area at a future time by utilizing the spatiotemporal characteristics of early warning events includes: Obtain the spatiotemporal feature vector of the early warning event Constructing a spatiotemporal sequence of historical early warning events , Indicates the number of warning events; Warning events for a future time. The conditional intensity function of the warning event is constructed using a Hawkes process. in, Indicates at time ,Location Location, type The probability of a warning event occurring; base strength ,in, Indicates the basic strength weight; Spatiotemporal basic features in, Indicates position Two-dimensional coordinates; Trigger function Among them, time-triggered core Indicates the time decay coefficient; Space Trigger Core covariance matrix It is used to control the spatial influence range and directionality; Indicates a warning event Within the affected area The variance of coordinates Indicates a warning event Within the affected area The variance of coordinates Represents the covariance coefficient; in, Indicates the type interaction coefficient; Severity adjustment items ,in, Indicates the severity amplification factor. Indicates an event The severity; Construct the likelihood function and solve for the parameters within the conditional strength function.

4. The multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to claim 3, characterized in that, The construction of the likelihood function and the solution of the parameters within the conditional strength function include: Construct the likelihood function: in, Indicates the scope of the target area; The likelihood function is solved using the gradient descent algorithm to obtain the likelihood parameters. .

5. The multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to claim 4, characterized in that, Also includes: For each historical warning event, Hawkes features are extracted, including: Background contribution ; Self-triggered contribution Interactive trigger contribution Target Area Early Warning Event Predicted trigger number Characteristics of the scope of impact of early warning events ; The above features are combined to obtain the Hawkes feature vector. ; For each historical early warning event, event-level features are extracted using a recurrent neural network (RNN), including: Encode each historical early warning event to obtain an event encoding vector. Among them, time coding Indicates the first The frequency of historical occurrences; Indicates type encoding; Periodic coding Spatial coding ;in, This represents a multilayer perceptron; Each historical early warning event is sorted by time and divided into time windows to obtain the spatiotemporal sequence of historical early warning events. ; Indicates the length of the time window; Indicates the number of time windows; The bidirectional long short-term memory (LSTM) algorithm is used to extract RNN feature vectors from the spatiotemporal sequence of historical early warning events. Among them, the bidirectional hidden state vector Hidden state one Hidden state two ; After aligning the Hawkes feature vector and the RNN feature vector on the time axis, they are then projected to a unified dimension to obtain the Hawkes feature vector with a unified spatiotemporal dimension. and RNN feature vectors ; in, Represents the Hawkes eigenmap matrix. Represents the feature mapping matrix of an RNN. and Represents the Hawkes feature correction vector and the RNN feature correction vector; use and Obtain a risk score for the target area at future moments.

6. The multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to claim 5, characterized in that, The use of and Obtain a risk score for the target area at future times, including: Through multiple algorithms and Integration, to obtain the corresponding integration features, including: By splicing, splicing integration features are obtained. ; By using gating fusion, we can obtain gating integration features. ;in, Indicates trainable gating parameters; Using integrated features to predict the risk of a target region includes: Divide the target area into multiple grids. , ,in, Indicates the number of grid cells; Predicting future time-point grid risk scores using ensemble vectors or , Represents the weight vector. Indicates the correction amount.

7. The multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to claim 6, characterized in that, Also includes: The method of guiding real-time acquisition of video data and prediction of public safety risks in target areas based on probability and historical early warning characteristics includes: Guided real-time acquisition of video data for the target area, including: Based on the conditional intensity function, a spatial risk heatmap of the target area is defined: weight value ; Based on RNN feature vectors, the risk time of the target region is defined. Represents the Fast Fourier Transform. Indicates attention weights. Represents the Dirac function; Calculate the spatial coverage priority of the camera, including: For cameras Let its coverage area be and ,in, Indicates the target area; Then the camera Spatial coverage priority in, For camera In position Observation quality weights; Calculate camera time scheduling priority , This indicates the calculation of the correlation coefficient. Indicates camera Historically, periods when high-risk events were captured on film; Set and adjust priorities for different event types. Then, the overall priority function in, , , Indicates priority weight. Indicates the event type weight; Based on the characteristics of the impact range of the early warning event, obtain the camera... optimal perspective Indicates a Gaussian distribution. , These represent the mean and variance of the adjustment of the camera's field of view relative to its original position, respectively. Based on the overall priority of the cameras, the corresponding cameras are scheduled first and their vision is adjusted to the optimal angle to collect video data of their coverage area in real time; the video data collected by multiple cameras constitutes the video stream data of the target area. Based on the acquired video stream data of the target area, public safety risk prediction is performed, including: Construct type interaction matrix Feature extractors are selected based on type interaction matrices, including: in, Indicates the threshold of clustering features. Individual interaction feature threshold, Crowd density feature extractor Represents a motion feature extractor. Indicates the target feature extractor. This represents a behavioral feature extractor; By combining appropriate feature extractors, relevant features are extracted from video stream data, input into a pre-trained early warning model, and output the probability of security risks.

8. The multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data according to claim 7, characterized in that, The process of extracting relevant features from video stream data through a combination of corresponding feature extractors, inputting these features into a pre-trained early warning model, and outputting a security risk probability includes: Constructed using 3D convolutional neural networks (I3D) or optical flow networks Extracting motion features of targets from video stream data ; The target recognition algorithm YOLO is used to construct Extracting the appearance features of targets from video stream data ; The CSRNet network for recognizing crowded scenes is constructed. Extracting crowd density features from video stream data ; The behavior of a target in video stream data is encoded using an autoencoder and a memory network, and the behavior anomaly score is output through the memory network. ; Will and The input is fed into a pre-trained group risk prediction model, which outputs the group risk probability. ; Will and The input is fed into a pre-trained individual risk prediction model, which outputs the probability of individual behavioral risk. ; if If so, it is determined that there is a risk to public safety. if If so, it is determined that there is an individual safety risk.

9. An electronic device, characterized in that, The electronic device is used to execute the multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data as described in any one of claims 1-8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the multi-algorithm collaborative public safety risk prediction method based on spatiotemporal and video perception data as described in any one of claims 1-8.

Images